Patient ventilator asynchrony detection

Abstract

A decision support tool is provided for identifying and assisting clinicians with patient ventilator asynchrony. The information used to make the identification may include data from a patient's ventilator including the volume, flow, and pressure associated with that ventilator. At least some of this information may be used to compute one or more features for a time series of the data received for the patient. These features may be used in connection with heuristic rules and machine learning algorithms to identify instances of patient ventilator asynchrony. Based on the identification, one or more intervening actions may be initiated to reduce the impact of patient ventilator asynchrony.

What is claimed is: 
     
       1. One or more non-transitory computer storage media having computer-executable instructions embodied thereon that, when executed on a computerized decision support system using one or more hardware processors, cause the one or more hardware processors to facilitate performance of a plurality of operations associated with a healthcare software program, the operations comprising:
 receiving a plurality of measurements taken by a ventilator associated with a patient, the plurality of measurements being acquired over a time span; 
 constructing one or more time series from the plurality of measurements for the ventilator to detect:
 (a) a plurality of peak measurement values using at least one of the one or more time series; and 
 (b) a plurality of breaths of the patient corresponding to the plurality of peak measurement values; 
 
 for each breath of the plurality of breaths, computing a plurality of features based on the plurality of measurements, the plurality of features selected from a group comprising at least:
 a flow ratio calculated from a ratio of exhalation volume and inhalation volume within an individual breath; 
 a volume oscillation ratio calculated from a ratio of a maximum volume oscillation and an in volume within an individual breath, wherein the maximum volume oscillation is calculated from a difference of minimum and maximum volume during a time when a volume is at least 70% of a maximum volume within an individual breath; 
 a volume peak time ratio calculated from a ratio between a volume peak time to next and a volume peak time to previous, wherein the volume peak time to next is calculated from a distance between a current breath's peak to the next breath's peak and the volume peak time to previous is calculated from a distance between the current breath's peak to a previous breath's peak; 
 a pressure peak time ratio calculated from a ratio between a pressure peak time to next and a pressure peak time to previous, wherein the pressure peak time to next is calculated from the time between a first peak of one breath to a first peak of a next breath and the pressure peak time to previous is calculated from the time between the first peak of one breath and a first peak in a previous breath; and 
 a time ratio calculated from a ratio of in time and out time within an individual breath; 
 
 detecting one or more asynchronies between the patient and the ventilator responsive to determining that the plurality of features correspond to one or more phenomena from a group comprising at least:
 a flow ratio less than 0.6; 
 a volume oscillation ratio greater than 0.25; 
 a volume peak time ratio less than 0.7; 
 a pressure peak time ratio less than 0.6; and 
 a time ratio less than 0.8; and 
 
 responsive to detecting the one or more asynchronies:
 automatically initiating an intervening action, 
 wherein:
 the intervening action comprises adjusting via the one or more hardware processors at least one of flow, volume, or pressure of the ventilator, and 
 the intervening action facilitates at least one effect from a group comprising:
 reducing a detected presence of the one or more asynchronies; 
 reducing an adverse effect associated with the one or more asynchronies; 
 countering the detected presence of the one or more asynchronies; and 
 countering the adverse effect associated with the one or more asynchronies. 
 
 
 
 
     
     
       2. The one or more non-transitory computer storage media of  claim 1 , wherein: the plurality of measurements comprise volume, pressure, and flow measurements; the one or more asynchronies comprise double triggering, auto triggering, premature cycling, or flow asynchrony; and the operations further comprise detecting one or more peaks within a series selected from a group comprising a volume time series, a pressure time series, and a flow time series. 
     
     
       3. The one or more non-transitory computer storage media of  claim 1 , wherein detecting the one or more asynchronies comprises applying a set of heuristic rules to the plurality of features for each breath to determine one or more anomaly groups that represent at least one asynchrony. 
     
     
       4. The one or more non-transitory computer storage media of  claim 3 , wherein detecting the one or more asynchronies further comprises applying one or more machine-learning electronic models to detect a likelihood of the one or more asynchronies being present. 
     
     
       5. The one or more non-transitory computer storage media of  claim 4 , wherein the set of heuristic rules is applied to a plurality of features before the one or more machine-learning electronic models are applied. 
     
     
       6. The one or more non-transitory computer storage media of  claim 1 , wherein constructing the one or more time series includes applying a smoothing algorithm, and wherein automatically initiating the intervening action comprises one or more operations selected from a group comprising electronically communicating a notification to a caregiver assigned to the patient, providing a recommendation to modify settings on the ventilator, and initiating an order for additional care for the patient. 
     
     
       7. The one or more non-transitory computer storage media of  claim 1 , wherein the operations further comprise after automatically initiating the intervening action:
 computing features associated with a subsequent breath based on additional measurements taken by the ventilator subsequent to the plurality of measurements; 
 detecting an asynchrony associated with the subsequent breath based on the features associated with the subsequent breath; and 
 initiating a second intervening action based on the asynchrony associated with the subsequent breath. 
 
     
     
       8. The one or more non-transitory computer storage media of  claim 7 , wherein the intervening action and the second intervening action comprise a first ventilator adjustment performed in response to the automatic initiating and a second ventilator adjustment performed after the first ventilator adjustment, respectively. 
     
     
       9. The one or more non-transitory computer storage media of  claim 1 , wherein detecting the one or more asynchronies is based on a machine-learning electronic model, and wherein:
 (a) the machine-learning electronic model is trained by inputting, to the machine-learning electronic model, vectors of data corresponding to instances of information associated with one or both of the plurality of features and the plurality of breaths, wherein the training of the machine-learning electronic model is based on learning weights associated with the instances of information, and 
 (b) detecting the one or more asynchronies comprises applying, to the trained machine-learning electronic model, information associated with data selected from a group comprising the plurality of features and the plurality of the breaths to facilitate the detection of the one or more asynchronies. 
 
     
     
       10. The one or more non-transitory computer storage media of  claim 1 , wherein the operations further comprise:
 automatically adjusting a ventilator setting in response to detecting the one or more asynchronies. 
 
     
     
       11. The one or more non-transitory computer storage media of  claim 10 , wherein the operations further comprise:
 after automatically adjusting the ventilator setting, (a) constructing one or more additional time series from additional measurements taken by the ventilator corresponding to a subsequent time span and (b) based on additional features computed from the additional measurements, detecting one or more additional asynchronies between the patient and the ventilator; and 
 automatically adjusting at least one of a group of ventilator settings based on the one or more additional asynchronies. 
 
     
     
       12. The one or more non-transitory computer storage media of  claim 11 , wherein the operations further comprise detecting an individual breath of the patient in response to identifying one or more peak measurement values in the one or more additional time series, and wherein the additional features are computed for the individual breath from the additional measurements. 
     
     
       13. The one or more non-transitory computer storage media of  claim 1 , wherein the operations further comprise:
 labeling one or more groups of anomalies, detected within each of the one or more time series, based on the plurality of features for each breath, wherein the one or more groups of anomalies comprise a type of asynchrony; 
 training a machine-learning electronic model, based on training data corresponding to the labeled groups of anomalies and the plurality of features; and 
 applying the machine-learning electronic model to information associated with the plurality of measurements to identify instances of asynchrony between the patient and the ventilator. 
 
     
     
       14. The one or more non-transitory computer storage media of  claim 1 , wherein the operations further comprise:
 training a machine-learning electronic model using training data corresponding to information associated with (a) instances of measurement data relating to ventilator and patient respiratory operations and (b) matches, mismatches, or asynchronies indicated by the instances of measurement data, wherein the training comprises performing a data transformation configured to reduce a dimensionality of the training data; and 
 inputting to the trained machine-learning electronic model one or more input sets of information to predict one or more asynchronies between the patient and the ventilator, wherein the one or more input sets of information are associated with the plurality of measurements and the plurality of features for each breath. 
 
     
     
       15. The one or more non-transitory computer storage media of  claim 14 , wherein the training comprises performing a data transformation configured to reduce a dimensionality of the training data. 
     
     
       16. The one or more non-transitory computer storage media of  claim 1 , wherein detecting one or more asynchronies comprises performing a data transformation configured to reduce a dimensionality of data associated with the plurality of measurements. 
     
     
       17. A computerized decision support system having one or more hardware processors configured to facilitate performance of a plurality of operations associated with a healthcare software application, the operations comprising:
 receiving a plurality of measurements taken by a ventilator associated with a patient, the plurality of measurements being acquired over a time span; 
 constructing one or more time series from the plurality of measurements for the ventilator to detect:
 (a) a plurality of peak measurement values using at least one of the one or more time series; and 
 (b) a plurality of breaths of the patient corresponding to the plurality of peak measurement values; 
 
 for each breath of the plurality of breaths, computing a plurality of features based on the plurality of measurements, the plurality of features selected from a group comprising at least:
 a flow ratio calculated from a ratio of exhalation volume and inhalation volume within an individual breath; 
 a volume oscillation ratio calculated from a ratio of a maximum volume oscillation and an in volume within an individual breath, wherein the maximum volume oscillation is calculated from a difference of minimum and maximum volume during a time when a volume is at least 70% of a maximum volume within an individual breath; 
 a volume peak time ratio calculated from a ratio between a volume peak time to next and a volume peak time to previous, wherein the volume peak time to next is calculated from a distance between a current breath's peak to the next breath's peak and the volume peak time to previous is calculated from a distance between the current breath's peak to a previous breath's peak; 
 a pressure peak time ratio calculated from a ratio between a pressure peak time to next and a pressure peak time to previous, wherein the pressure peak time to next is calculated from the time between a first peak of one breath to a first peak of a next breath and the pressure peak time to previous is calculated from the time between the first peak of one breath and a first peak in a previous breath; and 
 a time ratio calculated from a ratio of in time and out time within an individual breath; 
 
 detecting one or more asynchronies between the patient and the ventilator responsive to determining that the plurality of features correspond to one or more phenomena from a group comprising at least:
 a flow ratio less than 0.6; 
 a volume oscillation ratio greater than 0.25; 
 a volume peak time ratio less than 0.7; 
 a pressure peak time ratio less than 0.6; and 
 a time ratio less than 0.8; and 
 
 responsive to detecting the one or more asynchronies:
 automatically initiating an intervening action, 
 wherein:
 the intervening action comprises adjusting via the one or more hardware processors at least one of flow, volume, or pressure of the ventilator, and 
 the intervening action facilitates at least one effect from a group comprising:
 reducing a detected presence of the one or more asynchronies; 
 reducing an adverse effect associated with the one or more asynchronies; 
 countering the detected presence of the one or more asynchronies; and 
 countering the adverse effect associated with the one or more asynchronies. 
 
 
 
 
     
     
       18. The computerized decision support system of  claim 17 , wherein the asynchronies are detected using one or more machine learning algorithms including a uniform manifold approximation and projection (UMAP) algorithm. 
     
     
       19. The computerized decision support system of  claim 17 , wherein the asynchronies are detected using a uniform manifold approximation and projection (UMAP) model. 
     
     
       20. The computerized decision support system of  claim 17 , wherein the intervening action comprises automatically adjusting, by the one or more hardware processors, one or more ventilator settings associated with the patient based on a type of asynchrony associated with the one or more asynchronies. 
     
     
       21. The computerized decision support system of  claim 17 , wherein detecting the one or more asynchronies comprises applying a set of heuristic rules to the plurality of features for each breath to determine one or more anomaly groups that represent at least one asynchrony. 
     
     
       22. The computerized decision support system of  claim 17 , wherein the operations further comprise, based on the plurality of features for each breath, further detecting one or more instances of leakage. 
     
     
       23. The computerized decision support system of  claim 17 , wherein the operations further comprise:
 applying a first machine-learning electronic model of a first type, to information associated with the plurality of measurements, to select features for a potential feature pool; 
 applying a second machine-learning electronic model of a second type, to information associated with the potential feature pool, to select one or more final feature sets; and 
 applying a third type of machine-learning electronic model of a third type, to information associated with the one or more final feature sets, to generate a prediction indicative of the one or more asynchronies. 
 
     
     
       24. The computerized decision support system of  claim 23 , wherein: the first type corresponds to a gradient boosted tree model, the second type corresponds to an adaptive GLMnet, and the third type corresponds to a generalized linear model. 
     
     
       25. A computerized method for detecting patient ventilator asynchrony, the computerized method comprising:
 receiving a plurality of measurements taken by a ventilator associated with a patient, the plurality of measurements being acquired over a time span; 
 constructing one or more time series from the plurality of measurements for the ventilator to detect:
 (a) a plurality of peak measurement values using at least one of the one or more time series; and 
 (b) a plurality of breaths of the patient corresponding to the plurality of peak measurement values; 
 
 for each breath of the plurality of breaths, computing via one or more hardware processors a plurality of features based on the plurality of measurements, the plurality of features selected from a group comprising at least:
 a flow ratio calculated from a ratio of exhalation volume and inhalation volume within an individual breath; 
 a volume oscillation ratio calculated from a ratio of a maximum volume oscillation and an in volume within an individual breath, wherein the maximum volume oscillation is calculated from a difference of minimum and maximum volume during a time when a volume is at least 70% of a maximum volume within an individual breath; 
 a volume peak time ratio calculated from a ratio between a volume peak time to next and a volume peak time to previous, wherein the volume peak time to next is calculated from a distance between a current breath's peak to the next breath's peak and the volume peak time to previous is calculated from a distance between the current breath's peak to a previous breath's peak; 
 a pressure peak time ratio calculated from a ratio between a pressure peak time to next and a pressure peak time to previous, wherein the pressure peak time to next is calculated from the time between a first peak of one breath to a first peak of a next breath and the pressure peak time to previous is calculated from the time between the first peak of one breath and a first peak in a previous breath; and 
 a time ratio calculated from a ratio of in time and out time within an individual breath; 
 
 detecting one or more asynchronies between the patient and the ventilator responsive to determining that the plurality of features correspond to one or more phenomena from a group comprising at least:
 a flow ratio less than 0.6; 
 a volume oscillation ratio greater than 0.25; 
 a volume peak time ratio less than 0.7; 
 a pressure peak time ratio less than 0.6; and 
 a time ratio less than 0.8; and 
 
 responsive to detecting the one or more asynchronies:
 automatically initiating an intervening action, 
 wherein:
 the intervening action comprises adjusting via the one or more hardware processors at least one of flow, volume, or pressure of the ventilator, and 
 the intervening action facilitates at least one effect from a group comprising:
 reducing a detected presence of the one or more asynchronies; 
 reducing an adverse effect associated with the one or more asynchronies; 
 countering the detected presence of the one or more asynchronies; and 
 countering the adverse effect associated with the one or more asynchronies. 
 
 
 
 
     
     
       26. The computerized method of  claim 25 , wherein detecting the one or more asynchronies is based further on a semi-supervised machine-learning electronic model. 
     
     
       27. The computerized method of  claim 26 , wherein the semi-supervised machine-learning electronic model is trained using a plurality of hand-labeled samples as ground truth data. 
     
     
       28. The computerized method of  claim 25 , wherein the intervening action reduces the detected presence of the one or more asynchronies. 
     
     
       29. The computerized method of  claim 25 , wherein the intervening action is initiated based at least on a first asynchrony of the one or more asynchronies, and further comprising after automatically initiating the intervening action:
 detecting a second asynchrony in an additional breath based on additional measurements taken by the ventilator after the plurality of measurements; and 
 based on the second asynchrony, initiating a second intervening action. 
 
     
     
       30. The computerized method of  claim 29 , wherein the second intervening action reduces a detected presence of the second asynchrony. 
     
     
       31. The computerized method of  claim 25 , further comprising:
 performing a first ventilator adjustment, in response to the automatic initiating, based on the plurality of measurements; and 
 performing a second ventilator adjustment, after the first ventilator adjustment, based on additional measurements taken by the ventilator after the plurality of measurements.